We show that AI only works when the human dimension remains above speed.
By setting boundaries and preserving human judgment, technology can serve the whole.
AI between speed and wisdom: the human dimension as a firm boundary condition
Artificial intelligence accelerates processes and expands our capabilities, but without clear boundaries, efficiency quickly overtakes wisdom. The question is not what is possible, but what is necessary and desirable. Those who place the human dimension at the center design and use AI in a way that strengthens dignity, autonomy, and connection.
Introduction
AI has, in a short time, entered healthcare, the judiciary, education, and business operations. Algorithms analyze behavior, make predictions, and steer choices. The promise is real: faster, more accurate, more accessible. At the same time, something feels uneasy. Who determines how decisions are made? Which biases travel along in data and models? What happens to professional expertise when we outsource tasks to systems that appear faster and more convincing than we are?
The core issue is not a technical detail but a societal choice: how do we keep the human dimension leading in systems that, due to scale and computing power, can displace our capacity for judgment? This blog centers on one question: what do leaders, designers, and professionals need to do today to direct AI toward humanity and justice—so that innovation does not collide with the values that sustain work and society?
Tension that matters
Proposition
In the AI era, “more control through systems” is a temptation; what works is “more responsibility through people.” Efficiency without ethics is acceleration without direction.
Explanation
Algorithms optimize for what you ask them to optimize and for what you feed them. When the goal implicitly becomes speed or cost reduction, everything that is difficult to measure—human dignity, context, relational harm—receives too little weight. Bias in training data reinforces existing patterns: inequality becomes rationally packaged.
Meanwhile, psychological contagion takes place: when systems speak convincingly, we are more inclined to adopt their outcomes, even when the underlying assumptions are unclear. “The model must be right” signals a subtle shift of responsibility.
Effective leadership reverses this dynamic: make goals explicit, separate tool and decision, and clarify where the boundary lies between advice and judgment. In this way, wisdom is not overtaken by speed. This is not only a moral choice; leading frameworks explicitly require demonstrable human oversight, explainability, and pathways for redress.
Application
Define in advance which decisions will, in principle, remain in human hands and on the basis of which values. Agree on three explicit criteria for each application (for example: justice, proportionality, recoverability) and refer to them in every decision.
Ensure that “human in the loop” is not merely procedural but real: professionals must have time and language to deviate, and deviations should not be punished but examined. The EU AI Act will enforce this step by step in the coming years, with stricter requirements for high-risk systems and mandatory human oversight. (Entered into force on August 1, 2024; first prohibited practices applicable from February 2, 2025.)
Making the undercurrent visible
Beneath the technology move roles, loyalties, and projections. Executives may project excessive hope onto technology as a “neutral referee.” Development teams may feel loyalty to speed and elegance, users to convenience, regulators to certainty.
Without language for this undercurrent, pseudo-objectivity arises (“the system says so”) or defensiveness (“we can’t do anything about it”). Name who carries which responsibility, where interests collide, and which values outweigh metrics. This makes the conversation honest and prevents moral judgment from evaporating into a black box. In the judiciary, this has been made explicit in European principles: fundamental rights, non-discrimination, quality and safety, transparency, and user control.
Practical example
Situation
A hospital explores AI support for diagnostics. The model recognizes patterns in scans and provides recommendations. Physicians appreciate the speed but fear that doubt and context will receive less space.
Intervention
The hospital defines a clear objective: faster and better triage without loss of medical judgment. A decision framework is introduced with three fixed questions for each recommendation: what is the model’s certainty, what contextual information is missing, and what does this mean for risk and recoverability?
A deviation log is created in which physicians can record why they deviate and what they weigh differently. A monthly interdisciplinary meeting, including patient representation, discusses patterns, bias signals, and effects.
Effect
Diagnostic turnaround times decrease, but above all trust increases: physicians feel supported rather than replaced, and patients experience better-explained decisions. Deviations prove instructive and lead to concrete improvements in both model and process. This aligns with international healthcare ethics guidelines: technology supports, the human decides, with attention to justice, inclusion, and recovery.
A framework: the three B’s of the human dimension in AI
Purpose (Bedoeling), Boundaries (Begrenzing), and Proof Loop (Bewijslus).
- Purpose makes explicit which human good the technology serves—not “as fast as possible,” but for example “fair access to care” or “better substantiation of decisions.”
- Boundaries establish red lines: decisions that will not be automated, groups that receive extra protection, and contexts in which use is undesirable.
- Proof Loop closes the circle: we systematically test for bias and impact, provide objection and redress mechanisms, and visibly learn from deviations.
This lens operationalizes international principles: human-centered, transparent, accountable, and continuously risk-managed.
What do norms and supervision require—and what does that mean tomorrow?
Europe has adopted a comprehensive AI Regulation. Its core elements include risk-based requirements, documentation and logging, data quality, human oversight mechanisms, and transparency. The European Commission has established an AI Office for coordination and enforcement. Organizations working with AI in Europe must now ensure governance, documentation, and meaningful human oversight.
Beyond legislation, international frameworks offer guidance for daily practice. The NIST AI Risk Management Framework provides practical language and processes for risk identification, emphasizing transparency, validation, monitoring, and human-centered goals.
The OECD AI Principles (2019, updated 2024) provide anchor points: human-centered, robust, transparent, accountable, and safe. UNESCO’s Recommendation on the Ethics of Artificial Intelligence (2021) anchors human rights and societal impact.
Sector-specific guidance adds further detail. The WHO formulates six guiding principles for AI in healthcare. UNESCO (2023) provides global guidance for generative AI in education. In the judiciary, the Council of Europe’s CEPEJ Ethical Charter defines legitimacy and rule-of-law conditions.
Law and rights: where the boundary lies
The GDPR (Art. 22) grants individuals the right not to be subject solely to automated decision-making that produces legal effects or similarly significant impacts. This requires design choices: meaningful human oversight, explainability for the affected person, and genuine opportunities for objection and redress—not merely “human in the loop” on paper, but demonstrably in practice.
From principles to practice: documenting, testing, improving
Two practical tools make transparency and accountability workable: datasheets for datasets and model cards for models. Datasheets record origin, composition, assumptions, and limitations; model cards document performance, intended use, and fairness findings. This makes explanation to boards, teams, and stakeholders concrete and enables auditing.
Pay attention to the scale of language models
Large language models deliver impressive results but bring real risks: data leakage, plausible nonsense, reinforcement of bias, and ecological costs. Bender & Gebru et al. (2021) warn against “stochastic parrots”: convincing text without understanding. Recommendations include data documentation, clear usage boundaries, and research beyond the “bigger is better” path.
Governance and certification
If you want to structurally ensure that AI remains “under control,” a management system helps. ISO/IEC 42001:2023 introduces the AI management system (AIMS), defining roles, processes, monitoring, and continuous improvement. Not a checklist, but a way to align goals, risks, and improvement cycles with regulation.
Public legitimacy requires public dialogue
Legitimacy does not arise in the boardroom alone. Research by the Ada Lovelace Institute and The Alan Turing Institute shows that citizens want AI that is visibly fair, explainable, and testable; regulation increases comfort when it is tangible in lived experience. Organize participatory forms—citizen panels, professional groups, client councils—and make visible what you learn.
From reflection to movement
Which decisions in our organization are too important to automate, precisely because dignity, justice, or trust are at stake?
In the coming week, consciously observe three moments when you or your team use “the system” as an argument, and examine what that shorthand represents: time pressure, uncertainty, or unspoken interests.
For every new AI application, apply one decision check: can we explain in two sentences to a critical outsider how this strengthens the human dimension—and which boundary we have drawn where the system does not decide?
Conclusion
AI becomes directional where we give it direction. Choose one application today, name its purpose and its boundary, and start with a simple proof loop—so that speed and wisdom reinforce each other, and the human dimension is not a side issue but the boundary condition under which innovation is allowed to accelerate.
Bronnen (APA)
• Ada Lovelace Institute, & The Alan Turing Institute. (2023). How do people feel about AI?https://www.adalovelaceinstitute.org/wp-content/uploads/2023/06/Ada-Lovelace-Institute-The-Alan-Turing-Institute-How-do-people-feel-about-AI.pdf
• Ada Lovelace Institute, & The Alan Turing Institute. (2025). How do people feel about AI? 2025 survey findings.https://attitudestoai.uk/assets/documents/How-do-people-feel-about-AI-2025-Ada-Lovelace-Institute.pdf
• Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21), 610–623. https://doi.org/10.1145/3442188.3445922
• Council of Europe, CEPEJ. (2018). European Ethical Charter on the use of artificial intelligence in judicial systems and their environment. https://www.coe.int/en/web/cepej/cepej-european-ethical-charter-on-the-use-of-artificial-intelligence-ai-in-judicial-systems-and-their-environment
• European Union. (2016). Regulation (EU) 2016/679 (General Data Protection Regulation), Article 22. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng
• European Union. (2024). Regulation (EU) 2024/1689 (Artificial Intelligence Act). Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng
• ISO/IEC. (2023). ISO/IEC 42001:2023 — Information technology — Artificial intelligence — Management system.https://www.iso.org/standard/42001
• Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., … Gebru, T. (2019). Model cards for model reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency (FAT ’19)*, 220–229. https://doi.org/10.1145/3287560.3287596
• National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0) (NIST AI 100-1). https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf
• OECD. (2019/2024). OECD AI Principles. https://oecd.ai/en/ai-principles
• UNESCO. (2021). Recommendation on the ethics of artificial intelligence.https://www.unesco.org/en/articles/recommendation-ethics-artificial-intelligence
• UNESCO. (2023). Guidance for generative AI in education and research.https://unesco.org.uk/site/assets/files/10375/guidance_for_generative_ai_in_education_and_research.pdf
• World Health Organization. (2021). Ethics and governance of artificial intelligence for health: WHO guidance.https://www.who.int/publications/i/item/9789240029200
• Zuboff, S. (2019). The age of surveillance capitalism: The fight for a human future at the new frontier of power.PublicAffairs. https://www.hachettebookgroup.com/titles/shoshana-zuboff/the-age-of-surveillance-capitalism/9781610395694/
• Crawford, K. (2021). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press. https://yalebooks.yale.edu/book/9780300264630/atlas-of-ai/
• Hildebrandt, M. (2020). Law for computer scientists and other folk. Oxford University Press. https://academic.oup.com/book/33735
• Gebru, T., Morgenstern, J., Vecchione, B., Vaughan, J. W., Wallach, H., Daumé III, H., & Crawford, K. (2021). Datasheets for datasets. Communications of the ACM, 64(12), 86–92. https://dl.acm.org/doi/fullHtml/10.1145/3458723
Opmerking bij APA-stijl: waar mogelijk zijn DOI’s en officiële (primaire) publicatiepagina’s gebruikt; sommige internationale richtlijnen hebben geen DOI en worden daarom met de officiële webpagina of het PDF-dossier geciteerd.